See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/365832590
Visual Narratives of the Covid-19 pandemic
ArticleinJournal of Data Science Statistics and Visualisation · November 2022
DOI: 10.52933/jdssv.v2i7.64
CITATIONS
0
READS
144
2 authors:
Adalbert Franz Wilhelm
Constructor University Bremen gGmbH
85 PUBLICATIONS583 CITATIONS
SEE PROFILE
Susan Vanderplas
University of Nebraska at Lincoln
25 PUBLICATIONS174 CITATIONS
SEE PROFILE
All content following this page was uploaded by Susan Vanderplas on 07 December 2022.
The user has requested enhancement of the downloaded file.
MMMMMM YYYY, Volume VV, Issue II. doi: XX.XXXXX/jdssv.v000.i00
Visual narratives of the COVID-19
pandemic
Susan VanderPlas
Statistics Department
University of Nebraska Lincoln
United States
Adalbert F.X. Wilhelm
School for Business
Social and Decision Sciences
Jacobs University Bremen
Abstract
COVID-19 has sparked a worldwide interest in understanding the dynamic
evolution of a pandemic and tracking the effectiveness of preventive measures
and rules. For this reason, numerous media and research groups have produced
comprehensive data visualisations to illustrate the relevant trends and figures. In
this paper, we will look at a selection of COVID 19 data visualisations to evaluate
and discuss the currently established visualisation tools in terms of their ability to
provide a communication channel both within the data science team and between
data analysts, domain experts and a general interested audience. Although there
is no set catalogue of evaluation criteria for data visualisations, we will try to
give an overview of the different core aspects of visualisation evaluation and their
competing principles.
Keywords: exploratory data visualisation, logarithmic scales, visual comparisons, R.
1. Introduction
Over the past two years, several waves of COVID-19 infections caused by different mu-
tations of the SARS-CoV-2 virus have swept across the globe, claiming lives, stressing
healthcare systems, and changing many different facets of our day-to-day experience.
According to the WHO, almost 600 million confirmed cases and over 6.45 million deaths
2 Visual narratives of COVID-19
have been reported as of late August 2022
1
. The pandemic has generated enormous
interest in epidemiological data, its analysis and visualisation. From the beginning of
the pandemic, data on the number of infections and COVID-related deaths has been
published daily and made available to the public. Media, politicians and individuals
use this data to build their narratives about the pandemic, discuss its evolution, justify
the measures taken and discuss different prevention strategies against the spread of
the virus. These goals and their priorities were adapted as dynamically as the virus
mutated and the pandemic changed pace, but as with any other data visualisation, the
general principles remain the same; ensuring clear understanding by organising graphics
in such a way that the story of the data is told most effectively.
1.1. Data Journalism and the Media
In recent decades, many media organisations have established data teams that received
unprecedented attention and wide-ranging opportunities during the pandemic to show-
case their skills and abilities, not only in visualising data, but also in explaining their
data collection and data analysis strategies and methods. COVID-19 allowed these
data journalists to prove their utility in newsrooms, presenting data in digestible form
to the public in news articles. Hence data journalism will certainly be one of the bene-
ficiaries of the COVID-19 pandemic and has already become an important part of news
publishing, with COVID-19 delivering many excellent applications, often presented in
an interactive visual format on the web, such as dashboards (Koch 2021).
At the same time, we still see a lot of defective graphics disseminated and shared;
some that violate fundamental statistical and visualisation design principles such as
accuracy, relevance, timeliness, clarity, coherence, and reproducibility. These principles
have been laid out in numerous standards for statistical reporting in the application
areas, such as the ESS standard for quality reports (Commission 2020), the CONSORT,
PRISMA, CHEERs guidelines, and others (UK EQUATOR Centre 2022). The use of
numbers, their graphic representation and the systematic creation of charts and tables
from data in the mass media have been accompanied from the beginning by efforts
to improve the communication of quantitative and statistical information. And while
efforts continue to be made in this area, historical documents describe many of the
same problems we still struggle with today (American Statistical Association 1915;
Haemer 1949; Koverman 1961; Fienberg 1979; Hoffrage et al. 2000; Tufte 2001; Rosling
and Zhang 2011; Otava and Mylona 2020). In other words, while our technology for
recording, visually summarising, and communicating information distilled from data
has vastly improved over the past 100 years, we still face many of the same problems
when designing charts which communicate data accurately and effectively.
In other areas, such as visualisation of uncertainty, the pandemic presents new chal-
lenges for scientific communication. How do we effectively show the uncertainty in the
data amid flawed tests, new strains, changing diagnostic criteria, and incredible vari-
ability in global public health infrastructure? Moreover, how do we do this without
further decreasing public trust in institutions responsible for shepharding us through
the pandemic? While most designers chose to side-step these questions in the moment,
as we assess the pandemic in graphical form, we should also consider what was omitted
1
https://www.who.int/docs/default-source/coronaviruse/who_mou_30_august-2022.pdf
Journal of Data Science, Statistics, and Visualisation 3
from our visual renderings.
1.2. Effective Data Communication
A best practice for effective communication (graphical or otherwise) is to present in-
formation in an understandable way (Gigerenzer et al. 2007; Gigerenzer 2011), for
example by saying "one in ten" instead of 10%, or using absolute rather than relative
numbers. Presenting information in a visually appealing graphic form and highlight-
ing the most important points in "fact boxes" is one extremely effective way to engage
readers while focusing on key takeaway points. This approach was developed to in-
crease effective communication between stakeholders such as healthcare providers and
patients (Harding-Zentrum fur Risikokompetenz 2022). Such "fact boxes" combined
with the dynamic layer of the internet allowed for illustrative simulations of the spread
of the epidemic, for example as presented on 14 March 2020 in the Washington Post
with the title "Why outbreaks like coronavirus spread exponentially, and how to flatten
the curve" (Stevens 2020). Nevertheless, both the complexity of phenomena and the
"bipolarity" of statistical thinking remain a challenge. While human thinking tends
to simplify patterns, and political communication also prefers a simple cause-effect re-
lationship, real phenomena are often multivariate. Thus, in studying COVID-19 and
predicting its spread, it is important not only to consider symptomatology, disease inci-
dence and geographic distribution, population behaviour patterns, government policies,
and impacts on the economy, schools, people in nursing homes, and society at large,
but also to incorporate these into data analyses and communication of results. Associ-
ations observed in the data can often be caused by confounding variables. In addition,
much of the data comes from observational studies, which usually makes robust causal
attribution problematic. However, statisticians who point out these limitations are at
risk of having their statements pulled out to support one side in a polarised debate
(McConway and Spiegelhalter 2021).
Finally, it is critical that data communication take into account the intended audience.
Graphics created to communicate with public health officials should be different from
those created for the purpose of informing the public. Designers must carefully consider
the visual capabilities, attention constraints, and mathematical sophistication of the
audience when creating a chart, but in this era of instantaneous global communication,
charts designed for specific audiences may still reach a large population of viewers
with different capabilities. In this paper, we focus primarily on graphics which are
intended for mass consumption, if only to reduce the scope to something which can be
(non-exhaustively) covered in a paper.
1.3. Data visualisation Design and Testing
Having briefly discussed the media landscape surrounding the COVID pandemic and
goals for communicating about data effectively, we should also briefly discuss the choices
that go into creating visual representations of data intended to convey quantitative
information. While the overall goal of a data visualisation is to tell the story of the
data clearly and effectively, how this story is conveyed involves a series of design choices:
Which variables are shown, What limits are imposed, What visual forms are used to
represent the data? How will the chart (and any surrounding story, in journalism) be
4 Visual narratives of COVID-19
presented to the viewer? As the designer makes these choices, they have in mind an
audience, and the audience’s assumed characteristics affect the design choices as well:
language, education level, likely medium used to access the visualisation (computer,
physical newspaper, tablet, cell phone) all of these considerations also impact the
design of the chart and what ideas the designer considers during the creative process.
Once the basic form of the chart is determined, there are additional considerations
that may be specified by the stylistic conventions of different news organisations, but
may also be deliberately chosen for visual effect: colour palettes, fonts, spatial layouts,
and annotations may be used to highlight information, convey additional points of a
narrative, or provide helpful information about how to read an unfamiliar chart. While
each of these design choices could be (and have been) the subject of additional paper(s),
it is important to acknowledge that the graphics we analyse in this paper did not simply
pop into being they are the product of a series of decisions, some consciously made and
others imposed by convention or technological capabilities. We will highlight particular
choices which were effective (or not) throughout this paper, but we will also leave some
undiscussed for the sake of brevity.
Of course, the creative process does not only involve the creator. Most graphics that are
created for a purpose are tested, either formally or informally, before they are published,
even if the test results are never themselves published. Because so many of the charts
we highlight in this paper were discussed and disseminated via social media, in some
cases we can provide contemporary discussions of the design process and reactions to
the charts we highlight in this paper. Where such examples are not provided, that
should in no way be taken as evidence that the design and testing process was less
formal; it may only have been less public.
1.4. Scope
Our focus in this paper is to highlight, analyse, and discuss a selection of charts and
graphics which caught our attention during the initial, real-time experience of the pan-
demic. As we primarily consume media from western Europe and the United States,
our assessment is primarily limited to news outlets catering to these audiences; a ret-
rospective global analysis of COVID graphics would certainly be interesting, but is
beyond the scope of this work. While the charts we have selected were of personal
interest during the fear and uncertainty of the initial stages of the COVID pandemic,
in this paper we assess these same charts retrospectively, from a world where COVID
is an acknowledged fact of life. While it still has significant impacts on our day-to-
day lives, the existence of treatments, vaccines, and other preventative measures, as
well as the human tendency to adapt to changing circumstances mean that the novel
coronavirus which emerged in 2019 no longer dominates our consciousness in the same
way. Because of this perspective difference, we have enough distance from the initial
cataclysm to be able to assess the emotional impact of the graphics at the time and to
assess their utility retrospectively.
2. The global narrative
On 11 March 2020, WHO declared the outbreak of the novel coronavirus disease
Journal of Data Science, Statistics, and Visualisation 5
(COVID-19) a pandemic, and since that date at the latest, the global distribution of
the disease has been in the public eye. The spatial spread of the virus and the resulting
cases and deaths during this initial period were commonly visualised by choropleth
maps, see for example Figure 1 showing the total number of infections reported in each
country as of January 14, 2022. A central element of the narrative during March 2020
was the ubiquity of the disease and the accompanying global impact.
Choropleth maps are quite commonly used by media companies and governmental
organisations and it is easy to find good and bad examples of their usage. While
choropleth maps based on raw numbers of cases might look convincing and are an
obvious choice under the circumstances, the use of these charts ignores a number of
well-know caveats for statistical reporting and visualisation:
1. Absolute value unsuitability: As explained in (Monmonier 2005; Slocum et al.
2008; Speckmann and Verbeek 2010) among others, choropleth maps are funda-
mentally unsuitable for the representation of absolute numbers. Viewers tend to
integrate similarly coloured areas of the map unconsciously, and perceive choro-
pleths as representations of density. They also do not help to convey the desired
message as the absolute numbers of COVID-19 cases are strongly influenced by
the population size of the country as well as by the number of tests performed
and the accuracy of the recording and reporting system.
2. The area-bias: The visual impression is determined more by the colour and the
geographical area of the individual countries than by the number of COVID cases.
Since the countries of the world differ extremely in area, the visual assessment is
distorted, especially in the case of neighbouring countries with similar numbers
but different areas.
3. colour-scheme selection: Much research in visualisation is concerned with the
appropriate choice of colour schemes, see (Brewer et al. 1997; Müller et al. 2021).
The choice of a continuous scale or a categorical scale, the choice of a scale that
promotes the recognition of patterns or a scale that supports the filtering out of
specific map details, influences the quality of choropleth maps.
The issues with choropleth maps were highlighted in trade publications during the
initial stages of the pandemic (Field 2020), but these publications did not penetrate
the bureaucracy of public health officials suddenly tasked with data visualisation in
real time at scale.
Proportional symbol maps, see Figure 2 or graduated symbol maps place scaled symbols
or diagrams directly on the input map, often on the centroid of the regions. The symbol,
most commonly a disk or a square, is scaled such that its area corresponds to the data
value of the region. Proportional symbol maps sacrifice familiar shapes and topological
connectedness for equal visual weight (Gao et al. 2019); while this representation avoids
the area bias, proportional symbol maps are subject to under-estimation biases (Shim
and Son 2008) that may be of particular concern when visualising epidemic disease
spread. In addition, the visual impression of any map-based illustration depends on
the projection chosen for the underlying map; the examples in Figure 1 and Figure 2
have a western-centric bias.
6 Visual narratives of COVID-19
Figure 1: Choropleth map of the COVID-19 cases world-wide by country (World Health
Organization 2022).
Figure 2: Proportional symbol map showing the cumulated reported COVID-19 related
deaths adjusted for population world-wide by country (Alcantara et al. 2022).
Journal of Data Science, Statistics, and Visualisation 7
Another modification of the standard choropleth to address the area-bias problem is to
relax the spatial borders of a region in order to show magnitude using area (as well as
colour). Various algorithms can be used to modify geographic boundaries to maintain
spatial continuity, or shapes such as hexagons or squares can be used to represent geo-
graphic entities (as in Figure 3a); with either choice, geometry, topology, or both must
be sacrificed. While these methods are more popular in the academic visualisation
literature, including analysis of the spread of COVID-19 (Yalcin 2022), they were not
overly common in the media for global representations of COVID information. This
may be because cartograms require more visual effort on the part of the viewer; in
addition, news organisations may be reluctant to assume their readers have the geo-
graphic sophistication necessary to translate between the distorted map of the globe as
drawn on the cartogram, and the undistorted mental map of the globe used for mental
comparison purposes. A spatial dot density map might have allowed for analysis of
data across geographic boundaries, but this approach was not commonly used to show
global case counts, though it was highlighted as a useful option at the national level
(Field 2020), as shown in Figure 3b.
The narrative of the global perspective seems to be highly limited to the aspect of a
pandemic affecting the entire globe; none of the above visualisations intends to provide
a deeper insight into the spatial distribution of the phenomenon, neither within the
administratively motivated spatial borders nor across them. The use of maps to assess
the global pandemic more often focuses on the comparative aspects: How are we doing
as opposed to our neighbors? Which strategy to fight COVID-19 is better? We will
examine these topics more closely in Section 4.
Throughout the pandemic, both in choropleth maps and time series graphs, creators
have had to grapple with scale. When scales automatically adjust, mental comparisons
across time can become misleading, because the colour scale from yesterday does not
mean the same thing as it did today, as the range has expanded and the colours have
stayed the same. It is not that often that day-to-day design changes (or lack thereof)
cause fundamental problems with the interpretation of data visualisation, but during
the COVID 19 pandemic, this situation was both extremely common and had real
consequences; people would glance at the current COVID risk map and assume that
"red" meant the same thing that it did yesterday, despite the fact that the underlying
values had changed (Abiad 2020; Rubel 2020; Matthews 2020). One example of this
issue over two months of NY Times graphics is shown in Figure 8; the overall colour
scheme used changes once, but the variable displayed changes (at least) 4 times over
the course of a two month period.
This problem is more noticeable in choropleths and other charts which show case counts
based on colour scale alone than in time series charts, where we are more likely to expect
and notice a scale change, but it is still a fundamental issue across chart types and of
particular concern when there is no natural upper bound on the variable displayed.
Some designers leaned into this by arranging their graphics to highlight the fact that
cases were “off the charts", as in Figure 4, implicitly cuing readers that the situation
was exceptional and worthy of their attention.
Early in the pandemic, as case numbers grew exponentially without appropriate public
health measures to restrain them, the scale problem for choropleth maps was particu-
larly complex; over time either the colour mapping changes or the map becomes useless
8 Visual narratives of COVID-19
(a) Hexagonal Cartogram of COVID-19 cases in the United States by state (The Atlantic
2021).
(b) A dot-density map of COVID cases in China (Field 2020).
Figure 3: Modifications of choropleth maps which attempt to address the area bias of
choropleths in different ways. Both methods are applied at the national scale rather
than at global scale because of limitations with the visual form or data quality.
Journal of Data Science, Statistics, and Visualisation 9
Figure 4: Chart showing the increase in cases as a result of the B.1.617.2 variant in
regions of the UK, May 27, 2021. Notably, this chart arranges areas to accommodate
certain regions whose cases are “off the chart" in order to emphasize the exceptional
situation. (Burn-Murdoch 2021a)
due to compression against the upper limits of the scale, as in Panchadsaram (2020).
The magnitude of the problem also depends on whether the scale uses a set of bins
(a discrete mapping to a continuous variable) or uses a continuous colour mapping; in
some cases, outlets switched between the two options over time or changed the underly-
ing unit entirely. Interestingly, media outlets and municipal dashboards were subject to
criticism when the colour scales changed in response to increasing case counts (Boeck
2020) as well as when the colour scale did not change (Sandalow 2020), illustrating
the difficulty of designing visualisations amid competing constraints and a constantly
evolving pandemic.
3. The temporal narrative
Visualisations are typically created for a specific purpose, and during a pandemic it is
particularly critical to show the evolution of the situation over time, as there are new
developments on a daily basis. Typically, this means that cases, hospitalisations, and/or
deaths are plotted on the y-axis, with date shown on the x-axis, as in Figure 4 and
Figure 5. This not only allows individuals to compare the present to past case counts,
but also allows for forecasting and prediction of the future based on the current state.
Time series plots are the most natural choice for these goals, and over the course of the
pandemic, many different attempts have been made to show time series information
using different graphical forms.
The most obvious time series plot is also the most common; plotting cases against
date, often with a 7 or 14-day smooth to handle periodicity in case reporting due
to the work week, as shown in Figure 5. This allows for individuals to assess the
current situation and easily compare the current number of cases to past times for
which the individual has a direct reference for what to expect in impact to daily life.
10 Visual narratives of COVID-19
Figure 5: COVID monitoring summary charts on the Washington Post’s Coronavirus
page (Washington Post 2022a).
Figure 6: Some outlets provide contextual information to assist readers with interpret-
ing time series plots (Washington Post 2022b).
Often, as in Figure 5, these time series show multiple variables: cases, deaths, and
hospitalisations (and sometimes vaccinations), to provide a more comprehensive view
of the situation. This is particularly useful as there has been some decoupling of case
numbers and hospitalisations/deaths with the emergence of effective vaccinations and
strains which are more contagious but potentially also less severe. Many outlets also
provide additional charts designed to provide some immediate context to readers, as
shown in Figure 6.
However, there are some limitations with this presentation; if we are interested in
showing the change in counts over time, viewers must assess the slope of the line,
rather than its position. We know that slope judgments are much less accurate than
position (Cleveland and McGill 1987), even in relatively simple situations; when we
add in our ability to assess exponential growth, our perceptual accuracy is even more
suspect. This issue will be discussed further in Section 5 when we discuss log scales,
which are one solution to our misperception of exponential growth.
Yet another approach to showing the temporal component of the pandemic was to
leverage web graphics to show the temporal component through animation; this freed
up the x axis which in the default configuration would have been devoted to time and
allowed another variable to be displayed. While this approach was common on social
media, it was reserved for specific use-cases and more carefully selected variables by
professional media outlets. Figure 7 shows an example from the Financial Times which
uses the x-axis to show GDP per capita and the y-axis to show vaccination rate, bubble
size indicates a country’s population. With so many variables and ratios of variables,
these charts are complex to read and use, even though they are visually appealing.
Other time series representations were more abstract, sacrificing exact data represen-
tations for a higher-level summary comparing regions and points in time without the
burden of numerical calculations. Several versions from the New York Times are shown
Journal of Data Science, Statistics, and Visualisation 11
Figure 7: Animated bubble chart showing vaccination rate relative to per capita GDP
for each country (F.T. Visual and Data Journalism Team 2022).
in Figure 8. The advantage of these displays is that they are very simple and allow for
viewers to gain an intuitive understanding of the data, however, they do not present pre-
cise numerical information and even the colour scale is (perhaps intentionally) opaque
it would be extremely difficult to translate the visualisation into any sort of accurate
numerical estimate. In addition, by hiding these details from the user, it is very difficult
to identify when the underlying mathematical representation changes over time, as it
did several times during the summer of 2020. The use of a similar colour scheme as
well as similar geometric representations made it extremely difficult for users to identify
that the value represented had changed. Clearly, the purpose of these charts is not to
provide numerical precision, but rather a comparative assessment of the status of one
region relative to others as well as relative to previous time points.
One unique time series plot created a representation of the case counts in the frequency
domain; instead of plotting cases as they occurred, the chart instead shows line segments
from 0 to N in y, where N is a predefined number of cases; then the chart resets to
show the next angle. This produces a sense of intensity of the waves of the pandemic
over time as shown in Figure 9. Interestingly, this frequency representation can also be
easily transferred into an audio domain, providing access for visually impaired users as
well as a multimodal representation of the data for sighted users.
There are many different ways that designers can use time series data to provide ad-
ditional contextual information and facilitate comparisons. Figure 10 shows COVID
cases, positivity rate, and hospital admissions from South Africa’s Gauteng province,
which was one of the first areas to experience a large wave of the Omicron variant.
This chart provides context for the Omicron wave, showing that while infections are
occurring at a much faster rate than in previous waves, hospital admissions seem to
be approximately following previous trends. In addition, it is clear that the chart is
highlighting the Omicron variant from the colour selection: the red used to show the
Omicron numbers is a sharp contrast from the green and blue shades used to show the
first three waves of the pandemic. Notably, even though red and green in combination
are usually to be avoided for the sake of colourblind viewers, the shades of blue/green
variants which are chosen are light enough that the red stands out even to someone
12 Visual narratives of COVID-19
(a) June 16, 2020 (b) July 16, 2020 (c) July 18, 2020 (d) August 1, 2020
Figure 8: A series of screenshots of the NY Times state-by-state display showing evolv-
ing colour schemes and even metrics; initially, a discrete scale of two-week change in
cases is shown; then, a similar colour scheme is used with a different metric. Around
July 17, the colour scheme and variable shown changed to weekly cases per capita (from
"fewer" to "more"), and by August 14, the variable shown changed back again ("falling"
to "rising") while the colour scheme stayed the same.
Figure 9: Frequency-domain case count graphics. Here, each 1000 deaths are repre-
sented by a single line segment whose slope represents the time taken to reach the next
1000 deaths. This provides viewers with a sense of the pace of the epidemic, rather than
the raw case counts in more standard time series representations. This representation
can also be shown in the auditory domain, providing access to those who are typically
excluded from visualisations due to vision loss (Vuillemot 2020).
Journal of Data Science, Statistics, and Visualisation 13
Figure 10: COVID Cases, positivity rate, and hospital admissions from South Africa’s
Gauteng province, on a log scale (top) and linear scale (bottom). Successive waves are
overlaid, showing that the increase in cases due to the Omicron variant was sharper than
any previously measured increase, while hospitalisations were much more comparable
to previous waves (Burn-Murdoch 2021b).
with colour vision deficiency.
While traditionally, time series information has been limited to charts with time on
a linear x-axis, many charts manipulated time to highlight different aspects of the
pandemic. In some cases, these modifications provided large amounts of utility for a
relatively small increase in complexity; for example, John Burn-Murdoch’s decision to
index Financial Times COVID graphs to the days since the N th case or death. This
decision is a small modification of the basic time series plot, but has important conse-
quences in the chart’s readability as well as the visual emphasis for users: the graphs in
the Financial Times were designed to emphasize "... that there’s an inevitability about
how coronavirus spreads... even if there are only a few cases in your country today,
based on all the data we have, you will end up going along that path, the same path
that the likes of Italy and Spain have been on so tragically" (Hannen 2020). By placing
all countries on an even footing divorced from the temporal randomness of when the
first case appeared, the viewer can effectively compare the effect of government policies
and other interventions on the trajectory over time (Burn-Murdoch 2020a,b).
The New York Times intentionally violated typical design guidelines in a controversial
chart; the x-axis was wrapped around an Archimedean spiral to provide a sense of
14 Visual narratives of COVID-19
Figure 11: COVID cases in the United States, 2020-2022. The original graph from the
New York Times is on the left (Shaman 2022), and a reenvisioned version which is more
perceptually friendly is on the right (Shrestha 2022).
periodicity in the year-by-year evolution of case counts, as shown in Figure 11. The
original form of this chart is prone to the line-width illusion (VanderPlas and Hofmann
2015), in addition to the well-known problems with polar charts (Hofmann et al. 2012;
Waldner et al. 2020). A re-envisioned version created by Sourya Shrestha, aligns counts
on the outside of the spiral, and displays them as radial lines, mitigating some of
the issues with the line-width illusion by directly showing each line as its own entity,
facilitating direct comparisons.
In addition, the re-envisioned chart includes a reference line at 100K cases/day, which
allows the reader to compare the severity of different waves directly. Still, it is more
work to interpret and compare peaks on this chart than on a similar linear time series
chart with the same data, as the reader must assess line width and then do a mental
rotation and shift operation in order to compare with any other time period. By
design, this type of chart devotes less area to previous years of the pandemic, which
may decrease the amount of focus given to that data.
Other designers used the capabilities of web graphics to show the time component in the
data through animation. In some cases, this took the form of animated bubble plots,
while others dispensed with the traditional form of a line graph altogether, showing
rising case counts using a "bar chart race"; an animated series of bar charts over time
that show how case counts in each country are increasing. A typical example can be
found on YouTube (Stats On Clock 2022); a screenshot is shown in Figure 12. This type
of chart is effective in showing instantaneous changes in counts over days for countries
with an extremely high number of cases, but may make it difficult for the viewer to
process how cases are changing in countries that do not dominate case counts. By
Journal of Data Science, Statistics, and Visualisation 15
Figure 12: A screenshot from Stats on Clock’s bar chart race showing counts of COVID
cases over time throughout the pandemic (Stats On Clock 2022).
focusing viewer attention on changes in relative totals of case counts, the graphic tends
to hide steady, proportional increases in cases across the globe, making it somewhat
difficult to perceive large global trends relative to small changes in comparative case
counts.
As people grappled with the scale of the pandemic and the lives lost as a result, a
different set of data displays attempted to provide context to this loss over time. In
the New York times, this took the form of a story showing the accumulation of the first
100,000 deaths due to COVID in the US, with occasional short quotes from individuals’
obituaries to humanise the names (Barry et al. 2020). While this is not a typical time
series chart, as the user scrolled through the page the pace of the names increases,
showing the scale of the pandemic in a visceral way. A less sentimental dot-density
timeline by the New York Times about 9 months later chronicled the nation reaching
450,000 deaths; the difference between the two visualisations brings to mind the quote
oft attributed to Stalin: "One death is a tragedy; a million deaths is a statistic" (Quote
Investigator 2016).
A slightly different approach by the BBC’s Visual and Data Journalism team, shown
in Figure 13, displayed the COVID cases and deaths as a growing flower, with the stem
proportional to the number of cases and the flower petals showing the deaths over time
(The BBC Visual and Data Journalism team 2020). The visualisation comes with sound
as well, an attempt to make the pandemic an auditory experience as well as a visual
one. The form, animation, and auditory experience are all indicative of an emotional
appeal (Kostelnick 2016; D’Ignazio and Klein 2020); with the flower representation as
an explicitly recognized symbol of grief; it is clearly not about showing specific case
counts and deaths numerically.
4. The ranking narrative
"The plague is always the others." This is the short formula for dealing with infectious
16 Visual narratives of COVID-19
Figure 13: The BBC Visual and Data Journalism Team’s COVID flower, showing the
cases and deaths over time (The BBC Visual and Data Journalism team 2020).
Journal of Data Science, Statistics, and Visualisation 17
diseases from a historical perspective (Thiessen 2021). Closing the borders and re-
stricting access to the country became a popular means of controlling the spread of the
disease at several different points during the pandemic. Unsurprisingly, the prevailing
visual narrative focused on comparisons often fueled by political rivalry, historical de-
pendencies, recent withdrawal from supranational institutions or regional competitions.
An ongoing debate about the true extent of the dangers of COVID-19 and how best
to combat it, combined with daily availability and public access to data across admin-
istrative levels, fostered ongoing competition and the use of leaderboards to show how
bad case counts were somewhere else.
The bar charts used for this purpose (see top image in Figure 14) are straightforward
and provide a good way to show the “leading” countries or regions. Emphasis on spe-
cific countries is either preconfigured by the creator of the chart or can be modified
interactively in the online version by check boxes and drop-down menus. These bar
charts also allow easy adjustment of the count data to population size or other mean-
ingful standard. Streamgraphs (see lower image in Figure 14) are a special type of
stacked area graphs and became quite popular around 2008 when they were used by
the New York Times to visualise box office results of movies (Di Bartolomeo and Hu
2016). Their usefulness depends heavily on the clarity of the pattern, as well as design
choices such as the ordering of the different groups. Small variations in the proportions
of the groups are almost impossible to detect, but streamgraphs give a good indication
of which group is predominant at any given time, even though they are subject to the
line-width or sine illusion (VanderPlas and Hofmann 2015). When paired with the
bar charts showing excess mortality, the streamgraph provides temporal context to the
instantaneous information in the bar graph.
Often, these comparative charts were combined with policy discussions, with individuals
challenged to spot certain policy interventions on the case-count graphs of different
localities with different policies (Weiss 2020). Figure 15 shows a chart from an article
which appeared in The Federalist, a conservative outlet in the United States, suggesting
that mask mandates are ineffective because it is difficult to spot the impact of the mask
mandate on the overall trend of cases. Of course, mask mandates are but one component
of a much broader pandemic management strategy, but the goal of the chart is clear;
the reader is supposed to conclude that cases and masks are not associated. These
charts were so common on social media that media outlets ran stories identifying them
as misinformation (Reuters Staff 2020); clearly, comparative charts were effectively
deployed to misinform as well as to inform.
5. To log or not to log
As COVID cases grow quasi-exponentially while there are susceptible members of the
population (subject to the effectiveness of mitigation measures and testing availability),
it seems natural to use log scales to allow for more effective comparisons of slight changes
in case counts over time. In addition, log scales make it possible to compare regions
with different populations or infection rates in the same chart. As noted previously,
however, interpreting log scales requires levels of numerical sophistication that may
not be appropriate for the general public. Even researchers do not always read and
interpret log scales correctly (Menge et al. 2018); expecting the general public to do so
18 Visual narratives of COVID-19
Figure 14: Ordered death rates in countries and the temporal evolution of the death
toll share in major world regions (FT Visual & Data Journalism team 2021).
Journal of Data Science, Statistics, and Visualisation 19
Figure 15: Deaths per million people in New York, Sweden, Texas, and Georgia, as
shown in the Federalist (Weiss 2020). The goal of this comparison chart is to lead the
reader to conclude that masks are ineffective, as there is not a clear difference between
Texas and Georgia. New York is also shown (though not labeled as requiring or not
requiring masks), and Sweden is shown as well, despite not being a US state at all,
leading the critical reader to conclude that the time series shown in this chart have
been cherry-picked to make a rhetorical point.
is difficult under normal circumstances (Heckler et al. 2013) is difficult. When panic,
fear, uncertainty, and doubt about the situation are added to the mix, it is easy to
imagine that we become even worse when interpreting graphics.
One issue with assessing the use of log scales is that their effectiveness changes with
the stage of the pandemic and the amount and variety of data shown. Initially, log
scales were incredibly useful at showing case counts, because minimal mitigation mea-
sures were in place and the growth of case counts (or presumptive positive cases, in
absence of available testing) was fairly close to exponential. In addition, the use of log
scales allowed for the comparison of total cases across entities with large population
differences; in the US, we could compare cases in New York and California with cases
in Michigan and Washington, even though the populations (and corresponding case
counts) of Michigan and Washington are much lower than the populations (and case
counts) of either New York or California.
While log scales are not necessarily intuitive, many outlets tried to make the graphs
more intuitive by adding reference lines, as shown in Figure 17.
However, after the first wave of COVID, the issues with log scales became more appar-
ent; it was difficult to detect slight increases in case counts that indicated the beginning
of a new wave amid a background level of spread, as demonstrated in Figure 18. Di-
agonal reference lines from the origin were also less helpful, as the growth of cases or
deaths was no longer approximately exponential and varied over time; for these refer-
ence lines to be effective there would need to be a clear idea of when the case counts
started to increase exponentially, which is difficult to determine whilst in the thick of a
potential COVID wave. Occasionally graphs with manually drawn reference lines were
20 Visual narratives of COVID-19
Feb Mar Apr May Jun
1
100
10000
Date
Daily average cases (7 day average)
C
a
l
i
f
o
r
n
i
a
M
i
c
h
i
g
a
n
N
e
w
Y
o
r
k
W
a
s
h
i
n
g
t
o
n
Linear
Feb Mar Apr May Jun
0
2500
5000
7500
10000
Date
Figure 16: In the early stages of the pandemic, log scales allowed the comparison of
raw case counts in locations with vastly different population and case counts.
Figure 17: Reference lines to compare exponential growth rates of deaths in different
countries. This provides some additional context that may help individuals use log
scale data more successfully. This approach was first featured in the Financial Times,
but was quickly adopted by the New York Times, 91-DIVOC, and other outlets. Graph
from the Financial Times (March 23, 2020), image from Kosara (2020).
Journal of Data Science, Statistics, and Visualisation 21
Baseline: 15 cases
Baseline: 50 cases
Baseline: 200 cases
Log
Linear
0 10 20 30 0 10 20 30 0 10 20 30
1
10
100
0
100
200
300
Total Exponential Background
Figure 18: One problem with log scales is that if there is a background level of spread,
it can be hard to notice the introduction of an additional source of exponential spread.
Linear scales do not have this problem - the exponential source is noticeable very quickly
in the total line, but on the log scale it is much harder to discern when the exponential
source causes the total line to diverge from the background. In the top-right corner, it
is difficult to identify that there is an exponential increase in cases amid the baseline
of a uniform level of spread over time, even though the exponential source makes up
approximately 50% of the cases at the end of the time period shown.
made available, but usually only in a retrospective manner, as in Figure 19.
While log scales have their problems, linear scales are not immune from issues either. It
can be very difficult to adequately compare to past situations when looking at the full
time series of case counts. For example, in Figure 20, it is difficult to tell whether the
first wave of COVID cases in March 2020 had an increase as fast as that in January of
2021; it is even more difficult to compare the order-of-magnitude of change in case rate
growth of January 2021 relative to January 2022 when the more contagious omicron
variant became prevalent.
It is not clear that the use of log or linear scales during the COVID-19 pandemic had
a large effect on public opinion. Several studies were conducted in the early stages of
the pandemic (Romano et al. 2020; Sevi et al. 2020; Ryan and Evers 2020) and results
seem to suggest that while individuals have difficulty understanding log scale graphs,
these issues do not tend to affect their support for intervention measures, perhaps in
part because COVID-related news saturated the news and opinions were set outside of
information provided in the experiments. This saturation makes it difficult to study
the graphical influences while removing effects of popular opinion, political leaning, and
emotional sentiment relating to the pandemic.
If we evaluate the use of log and linear scales under the more general context of ex-
ponential (or near-exponential) growth rates, we can gain some clarity as to their use
in this particular context. We are abysmal at forecasting exponential growth, vastly
under-estimating future growth by using linear or quadratic approximations (Wagenaar
and Timmers 1978; Lawrence and O’Connor 1992; Timmers and Wagenaar 1977). If
22 Visual narratives of COVID-19
Figure 19: An annotated time series from the Financial Times which appeared on
Twitter on November 12, 2020. This chart has annotations which show the decreasing
R
eff
, the effective rate of exponential spread, in successive peaks of the pandemic, with
periods of R
eff
< 1 between surges in cases. These annotations assist the reader with
drawing complex conclusions from the chart, but are difficult to automate and thus
tend to be manually curated (Gubrud 2020).
0
20000
40000
60000
2020−07 2021−01 2021−07 2022−01 2022−07
Date
Daily Cases reported (7−day average)
Reported COVID cases in New York State, 2020−2022
Figure 20: Reported COVID cases in New York State, 2020-2022. The linear scale
makes it difficult to compare the trajectory of different waves to determine how severe
the current status is relative to the past, because the primary contrast is the height of
the relative peaks, rather than the growth rate. A similar graph on the log scale would
have the peaks at much more similar heights (though there would still be a difference),
allowing the reader to focus on other information, such as the slope of the relative lines.
Journal of Data Science, Statistics, and Visualisation 23
the goal of a chart of COVID case counts is to allow individuals to forecast the trend
and make decisions accordingly, then using a log scale (with all of its pitfalls in under-
standing) may be the best option, as it at least replaces the need to forecast along an
exponential curve with the need to forecast along a straight line. However, it is not
clear that individuals can transfer that prediction back to a linear scale. If the goal
of presenting a chart of case counts is to tell the story of what has happened (rather
than supporting future decision-making), then it is undoubtedly better to present the
data on a more familiar linear scale that requires less cognitive load on the part of the
viewer.
Unfortunately, there are not a lot of good options for supporting forecasting decisions
in mathematically unsophisticated viewers. Guide lines, like those used in the Financial
Times, were helpful in the initial phases of the pandemic, but no outlets that we are
aware of shifted the lines to show the growth rate of the current peak (as opposed to
the initial peak), and due to the issues shown in Figure 18, these guide lines may not
have been all that successful in any case. Those hoping to create successful time series
charts that allow forecasting are thus left with two options: deal with our inability to
forecast exponential growth, or deal with most individuals’ inability to translate log
scales into practical reality.
6. Summary and Discussion
Throughout the pandemic (thus far), visual representations of data have been an inte-
gral part of scientific communication. While not always optimal, these graphics have
attempted to provide meaningful context, to encourage individual and collective ac-
tion, and to help individuals grapple with the scale of the pandemic in cases, deaths,
vaccinations, and interventions. As in any developing situation, choices made at some
points in the pandemic did not always persist many news outlets refined their charts
and approach to the design of graphics over the course of the pandemic. This evolution
of graphical forms has provided us with an opportunity to evaluate what worked and
did not and why in a context of broad general interest.
The visual narratives of the pandemic which have persisted over time are due in part
to the rise of data journalism and interactive graphics platforms which ensure that
every outlet has the ability to host engaging and visually appealing graphics; however,
while these graphics are nice to look at, not all are equally functional or useful for
communication purposes. Thus, analysing the rich collection of visualisations produced
as part of the pandemic provides a unique opportunity to examine how these graphics
were (and were not) effective, as well as the design decisions made during their creation.
As in any situation, good tools can be used and misused freely; we have seen charts
and graphics used to perpetuate misinformation and misleading claims, even though it
is far more common for charts and graphs to be used for good to educate and inform
the public.
The COVID 19 pandemic has amply demonstrated that policies are only effectively
implemented and followed by the population if they are accepted by a large majority
of the population. To achieve this goal, it is essential that scientists, governments, the
media, and citizens all effectively communicate in order to justify the appropriateness,
24 Visual narratives of COVID-19
usefulness and relevance of the measures. This is as true for guidelines about the use
and creation of charts as it is for policies like masks and vaccines.
The pandemic and graphics used to show the ebb and flow of COVID cases highlight the
challenges of data visualisation in the modern era. Even with numerous guidelines and
standards for graphics, it is not straightforward to communicate complex data; creating
charts which fully represent the intended message or messages remains a challenge.
It is essential that we continue to study how graphics are perceived and interpreted,
supporting the guidelines and standards which exist with user studies that ground these
rules in science.
Journal of Data Science, Statistics, and Visualisation 25
Computational Details
Graphics included in this paper not sourced from media outlets were created using data
published by the New York Times at https://github.com/nytimes/covid-19-data/.
Graphics were created using ggplot2 (Wickham 2016) and R 4.1.2. R itself and all
packages used are available from the Comprehensive R Archive Network (CRAN) at
https://CRAN.R-project.org/.
Archived data and code to create the figures in this paper can be found at https:
//github.com/srvanderplas/covid-visualization-dssv.
Archived Information
This paper references many news articles which contain interactive graphics. In order
to preserve these graphics (which depend on a great many external files), the authors
have preserved many of the interactive graphics as self-contained html files (Vanderplas
2022). The authors have also made extensive use of the Internet Wayback Machine,
which aims to preserve the state of important web pages as they change over time.
Acknowledgments
First ideas have been presented at DSSV 2021. We thank our colleagues, the reviewers,
and the editors for valuable feedback and encouragement to complete and revise this
manuscript.
References
Abiad, F. (2020). A color diagram by @WHOLebanon for the distribution of Covid
cases by district (September 21 Vs. 25). The rise and spread of cases was such that the
scale had to be changed. The graph on the right actually shows worsening numbers
though the change in the color can be misleading. https://t.co/IZKVi9ePjH. https:
//twitter.com/firassabiad/status/1310202386821132293.
Alcantara, C., Shin, Y., Shapiro, L., Taylor, A., and Emamdjomeh, A. (2022). Map-
ping the worldwide spread of the coronavirus. https://www.washingtonpost.com/
graphics/2020/world/mapping-spread-new-coronavirus/.
American Statistical Association (1915). Joint committee on standards for graphic
presentation. Publications of the American Statistical Association, 14(112):790–
797, ISSN: 15225437, DOI: 10.2307/2965153, http://www.jstor.org/stable/
2965153. tex.ids= JointCommitteeStandards1915 publisher: [American Statistical
Association, Taylor & Francis, Ltd.].
Barry, D., Buchanan, L., Cargill, C., Daniel, A., Delaquérière, A., Gamio, L., Gianor-
doli, G., Harris, R., Harvey, B., Haskins, J., Huang, J., Landon, S., Love, J., Maalouf,
G., Matthews, A., Mohamed, F., Moity, S., Royal, D.-C., Ruby, M., and Weingart,
26 Visual narratives of COVID-19
E. (2020). Remembering the 100,000 Lives Lost to Coronavirus in America. The New
York Times, ISSN: 0362-4331, https://www.nytimes.com/interactive/2020/05/
24/us/us-coronavirus-deaths-100000.html.
Boeck, T. (2020). #DataViz decisionmaking in action: was confused why the NYT
COVID 19 map was showing many states lighter than I remembered (lighter =
fewer cases). Upon closer inspection, they’ve been forced to change their color scale
due to exploding case numbers. https://t.co/p9Le8eTIQx. https://twitter.com/
toriboeck/status/1329327634619228162.
Brewer, C. A., MacEachren, A. M., Pickle, L. W., and Herrmann, D. (1997). Mapping
mortality: Evaluating color schemes for choropleth maps. Annals of the Association
of American Geographers, 87(3):411–438, http://www.jstor.org/stable/2564061.
Burn-Murdoch, J. (2020a). All countries’ death toll trajectories: Increasingly
looks like India tracking S Korea, Australia’s curve tracking Japan, 38 countries
now shown here, more every day. https://twitter.com/jburnmurdoch/status/
1244721960946409472/photo/1.
Burn-Murdoch, J. (2020b). Latest case trajectories for major countries: US
cases continue to soar. unquestionably the worst outbreak in the world. re-
porting from @yuanfenyang & co on whether you can trust the Chinese data:
https://ft.com/content/4aa35288-3979-44f7-b204-b881f473fca0. https://twitter.
com/jburnmurdoch/status/1244722153041387520.
Burn-Murdoch, J. (2021a). NEW: B.1617.2 is fuelling a third wave in the UK, with
not only cases but also hospital admissions rising. Vaccines will make this wave
different to those that have come before, but it remains a concern, and one that other
countries will soon face. Thread on everything we know: https://t.co/4825qOqgrl.
https://twitter.com/jburnmurdoch/status/1397995388267810818.
Burn-Murdoch, J. (2021b). NEW: today’s update from Gauteng, now on a log scale to
better show current trajectories. Steepness of lines shows how much faster the growth
in cases and positivity is now vs past waves, and hospital admissions are now steep-
ening too as the acceleration in cases feeds through. https://t.co/xBNGzxmwhC.
https://twitter.com/jburnmurdoch/status/1466480113487392769.
Cleveland, W. S. and McGill, R. (1987). Graphical Perception: The Visual Decod-
ing of Quantitative Information on Graphical Displays of Data. Journal of the
Royal Statistical Society. Series A (General), 150(3):192, ISSN: 00359238, DOI:
10.2307/2981473, https://www.jstor.org/stable/2981473?origin=crossref.
Commission, E. (2020). ESS Standard for Quality Reference Line. https://unstats.
un.or-g/unsd/dnss/docs-nqaf/eurostat-esqr_final.pdf.
Di Bartolomeo, M. and Hu, Y. (2016). There is more to streamgraphs than movies:
Better aesthetics via ordering and lassoing. Comput. Graph. Forum, 35(3):341–350,
ISSN: 0167-7055.
Journal of Data Science, Statistics, and Visualisation 27
D’Ignazio, C. and Klein, L. F. (2020). Data Feminism. Ideas Series. The MIT Press,
Cambridge, Massachusetts, ISBN: 978-0-262-35852-1.
Field, K. (2020). Mapping coronavirus, responsibly. https://www.esri.com/
arcgis-blog/products/product/mapping/mapping-coronavirus-responsibly/.
Fienberg, S. E. (1979). Graphical Methods in Statistics. The American Statisti-
cian, 33(4):165–178, ISSN: 0003-1305, DOI: 10/chmrvh, https://www.jstor.org/
stable/2683729.
FT Visual & Data Journalism team (2021). Coronavirus tracker: the latest figures
as countries fight the Covid-19 resurgence. Financial Times, https://www.ft.com/
content/a2901ce8-5eb7-4633-b89c-cbdf5b386938.
F.T. Visual and Data Journalism Team (2022). Covid-19 vaccine tracker: the global
race to vaccinate. https://ig.ft.com/coronavirus-vaccine-tracker/.
Gao, P., Li, Z., and Qin, Z. (2019). Usability of value-by-alpha maps com-
pared to area cartograms and proportional symbol maps. Journal of Spatial Sci-
ence, 64(2):239–255, ISSN: 1449-8596, DOI: 10.1080/14498596.2018.1440649,
https://doi.org/10.1080/14498596.2018.1440649. Publisher: Taylor & Fran-
cis _eprint: https://doi.org/10.1080/14498596.2018.1440649.
Gigerenzer, G. (2011). What are natural frequencies? BMJ, 343(oct17 2):d6386–d6386,
ISSN: 0959-8138, 1468-5833, DOI: 10.1136/bmj.d6386, https://www.bmj.com/
lookup/doi/10.1136/bmj.d6386.
Gigerenzer, G., Gaissmaier, W., Kurz-Milcke, E., Schwartz, L. M., and Woloshin,
S. (2007). Helping Doctors and Patients Make Sense of Health Statistics. Psy-
chological Science in the Public Interest, 8(2):53–96, ISSN: 1529-1006, DOI:
10.1111/j.1539-6053.2008.00033.x.
Gubrud, M. (2020). @DrEricDing We’ve basically cycled a small fraction of the reduc-
tion we had achieved, a smaller fraction in the 3rd surge than the second. Again,
this is evidence that more use of better masks, ventilation, testing, etc. can make
the difference & drive the pandemic into steep decline. https://t.co/x9NoOgRDA7.
https://twitter.com/mgubrud/status/1326784082399944704.
Haemer, K. W. (1949). Presentation Problems: Area Bias in Map Presenta-
tion. The American Statistician, 3(2):19–19, ISSN: 0003-1305, 1537-2731, DOI:
10.1080/00031305.1949.10483438, http://www.tandfonline.com/doi/abs/10.
1080/00031305.1949.10483438.
Hannen, T. (2020). Coronavirus trajectory tracker explained. https://www.ft.com/
video/9a72a9d4-8db1-4615-8333-4b73ae3ddff8.
Harding-Zentrum fur Risikokompetenz (2022). Faktenboxen zur mRNA-Schutzimpfung
gegen COVID-19. https://www.hardingcenter.de/de.
28 Visual narratives of COVID-19
Heckler, A. F., Mikula, B., and Rosenblatt, R. (2013). Student accuracy in reading log-
arithmic plots: The problem and how to fix it. In 2013 IEEE Frontiers in Education
Conference (FIE), pages 1066–1071. DOI: 10.1109/FIE.2013.6684990.
Hoffrage, U., Lindsey, S., Hertwig, R., and Gigerenzer, G. (2000). Communicat-
ing statistical information. Science, 290(5500):2261–2262, ISSN: 0036-8075, DOI:
10.1126/science.290.5500.2261, https://science.sciencemag.org/content/
290/5500/2261.
Hofmann, H., Follett, L., Majumder, M., and Cook, D. (2012). Graphical tests for power
comparison of competing designs. IEEE Transactions on Visualization and Computer
Graphics, 18(12):2441–2448, DOI: https://doi.org/10.1109/TVCG.2012.230.
Koch, T. (2021). Welcome to the revolution: COVID-19 and the democ-
ratization of spatial-temporal data. Patterns, 2(7), ISSN: 2666-3899, DOI:
10.1016/j.patter.2021.100272, https://www.cell.com/patterns/abstract/
S2666-3899(21)00098-2.
Kosara, R. (2020). In Praise of the Diagonal Reference Line. https://eagereyes.
org/blog/2020/in-praise-of-the-diagonal-reference-line.
Kostelnick, C. (2016). The Re-Emergence of Emotional Appeals in Interactive Data
Visualization. Technical Communication, 63(2):116–135.
Koverman, H. C. (1961). PLANNING THE VISUAL PRESENTATION. The
Professional Geographer, 13(6):24–26, ISSN: 0033-0124, 1467-9272, DOI:
10.1111/j.0033-0124.1961.136_24.x, http://www.tandfonline.com/doi/abs/
10.1111/j.0033-0124.1961.136_24.x.
Lawrence, M. and O’Connor, M. (1992). Exploring judgemental forecasting.
International Journal of Forecasting, 8(1):15–26, ISSN: 0169-2070, DOI:
10.1016/0169-2070(92)90004-S, http://www.sciencedirect.com/science/
article/pii/016920709290004S.
Matthews, C. (2020). A lesson in futzing with data to make a country seem less covid
infected: just change the # scale of what your colors mean. As if by magic, things
look better! Until someone like me comes along and redraws the map in photoshop
using last week’s color scale. cc @markrutte @rivm https://t.co/Oil1IOGJP6. https:
//twitter.com/matthewscd/status/1315604109819551749.
McConway, K. and Spiegelhalter, D. (2021). Sound human, steer clear of jargon, and
be prepared. Significance, 18(2):32–34, DOI: 10.1111/1740-9713.01508.
Menge, D. N. L., MacPherson, A. C., Bytnerowicz, T. A., Quebbeman, A. W.,
Schwartz, N. B., Taylor, B. N., and Wolf, A. A. (2018). Logarithmic scales
in ecological data presentation may cause misinterpretation. Nature Ecology &
Evolution, 2(9):1393–1402, ISSN: 2397-334X, DOI: 10.1038/s41559-018-0610-7,
http://www.nature.com/articles/s41559-018-0610-7.
Monmonier, M. (2005). Lying with maps. Statistical Science, 20(3):215–222, ISSN:
08834237, http://www.jstor.org/stable/20061176.
Journal of Data Science, Statistics, and Visualisation 29
Müller, A., Lausser, L., Wilhelm, A., Ropinski, T., Platzer, M., Neumann, H., and
Kestler, H. A. (2021). A perceptually optimised bivariate visualisation scheme for
high-dimensional fold-change data. Advances in Data Analysis and Classification,
15(2):463–480, ISBN: 1862-5355, DOI: 10.1007/s11634-020-00416-5, https://
doi.org/10.1007/s11634-020-00416-5.
Otava, M. and Mylona, K. (2020). Communicating statistical conclusions
of experiments to scientists. Quality and Reliability Engineering Interna-
tional, 36(8):2688–2698, DOI: https://doi.org/10.1002/qre.2697, https://
onlinelibrary.wiley.com/doi/abs/10.1002/qre.2697.
Panchadsaram, R. (2020). If we are at war with this virus, we need to push away parti-
sanship and attack it with the interventions that work. **Things are getting worse.**
We just updated http://covidexitstrategy.org and added a new color, "Bruised Red"
to capture the new extremes that we are seeing. https://twitter.com/rypan/
status/1282020934044401664.
Quote Investigator (2016). A Single Death Is a Tragedy; A Million Deaths Is a Statistic.
https://quoteinvestigator.com/2010/05/21/death-statistic/.
Reuters Staff (2020). Fact check: Mask mandates are not causing a
spike in COVID-19 infections. Reuters, https://www.reuters.com/article/
uk-factcheck-mask-mandates-n0t-causing-c-idUSKBN2802W4.
Romano, A., Sotis, C., Dominioni, G., and Guidi, S. (2020). The scale of COVID-
19 graphs affects understanding, attitudes, and policy preferences. Health Eco-
nomics, 29(11):1482–1494, ISSN: 1099-1050, DOI: 10.1002/hec.4143, https://
onlinelibrary.wiley.com/doi/abs/10.1002/hec.4143.
Rosling, H. and Zhang, Z. (2011). Health advocacy with gapminder animated statis-
tics. Journal of Epidemiology and Global Health, 1:11–14, ISSN: 2210-6014, DOI:
https://doi.org/10.1016/j.jegh.2011.07.001, https://doi.org/10.1016/j.
jegh.2011.07.001.
Rubel, L. (2020). hey @UpshotNYT i’m curious about your COVID choropleth maps?
as the data changes, do you shift the scale so to maintain five 5 color categories?
In other words, is today’s red tomorrow’s orange? i don’t know how to see old
maps to figure it out myself #maps #dataviz https://t.co/nCYagMS6ip. https:
//twitter.com/Laurie_Rubel/status/1327982127749722112.
Ryan, W. H. and Evers, E. R. K. (2020). Graphs with logarithmic axes distort
lay judgments. Behavioral Science & Policy, 6(2):13–23, ISSN: 2379-4615, DOI:
10.1353/bsp.2020.0011, https://muse.jhu.edu/article/799814.
Sandalow, S. (2020). ChiTribGraphics please update the color scale on your COVID
by ZIP map. Practically every zip in the city has 500 cases. https://twitter.com/
lastyear/status/1290835141825265664.
Sevi, S., Aviña, M. M., Péloquin-Skulski, G., Heisbourg, E., Vegas, P., Coulombe, M.,
Arel-Bundock, V., Loewen, P. J., and Blais, A. (2020). Logarithmic versus Linear
30 Visual narratives of COVID-19
Visualizations of COVID-19 Cases Do Not Affect Citizens’ Support for Confinement.
Canadian Journal of Political Science, 53(2):385–390, ISSN: 0008-4239, 1744-9324,
DOI: 10.1017/S000842392000030X.
Shaman, J. (2022). Opinion | Here’s When We Expect Omicron to
Peak. The New York Times, https://www.nytimes.com/2022/01/06/opinion/
omicron-covid-us.html.
Shim, J.-B. and Son, I. (2008). An Analysis of the Middle School Students’ Abilities
to Recognize the Proportional Symbol Maps. Journal of the Korean Geographi-
cal Society, 43(4):638–654, ISSN: 1225-6633, https://koreascience.kr/article/
JAKO200801440605855.page. Publisher: The Korean Geographical Society.
Shrestha, S. (2022). Covid spiral. https://github.com/souryashrestha/covid_
spiral.
Slocum, T., McMaster, R., Kessler, F., Howard, H., and Mc Master, R. (2008). The-
matic Cartography and Geographic Visualization. Prentice Hall, 3rd edition.
Speckmann, B. and Verbeek, K. (2010). Necklace maps. IEEE Transactions on Visu-
alization and Computer Graphics, 16(6):881–889, DOI: 10.1109/TVCG.2010.180.
Stats On Clock (2022). Coronavirus Cases Bar Chart Race (Till 2022, JAN ) |
COVID-19 Spread Around the World | SOC. https://www.youtube.com/watch?
v=8WH-bTH6GqI.
Stevens, H. (2020). These simulations show how to flatten the coronavirus growth curve.
https://www.washingtonpost.com/graphics/2020/world/corona-simulator/.
The Atlantic (2021). State Cases per 1M People. https://covidtracking.com/data/
charts/us-daily-positive-cases.
The BBC Visual and Data Journalism team (2020). Coronavirus: How can we imag-
ine the scale of Covid’s death toll? https://www.bbc.co.uk/news/resources/
idt-7464500a-6368-4029-aa41-ab94e0ee09fb. Section: uk.
Thiessen, M. (2021). Auf Abstand: Eine Gesellschaftsgeschichte der Coronapandemie.
Campus Verlag.
Timmers, H. and Wagenaar, W. A. (1977). Inverse statistics and misperception of
exponential growth. Perception & Psychophysics, 21(6):558–562, ISSN: 1532-5962,
DOI: 10.3758/BF03198737, https://doi.org/10.3758/BF03198737.
Tufte, E. R. (2001). The Visual Display of Quantitative Information. Graphics Press,
Cheshire, CT, 2 edition.
UK EQUATOR Centre (2022). The EQUATOR Network Enhancing the QUAlity and
Transparency Of Health Research. https://www.equator-network.org/.
Vanderplas, S. (2022). COVID-19 data graphics archive. DOI:
10.6084/m9.figshare.21152971.v1.
Journal of Data Science, Statistics, and Visualisation 31
VanderPlas, S. and Hofmann, H. (2015). Signs of the Sine Illusion—Why We Need to
Care. Journal of Computational and Graphical Statistics, 24(4):1170–1190, ISSN:
1061-8600, DOI: 10.1080/10618600.2014.951547, https://doi.org/10.1080/
10618600.2014.951547.
Vuillemot, R. (2020). Sonification of COVID-19 data / Romain Vuillemot / Observable.
https://observablehq.com/@romsson/sonification-of-covid-19-data.
Wagenaar, W. A. and Timmers, H. (1978). Extrapolation of exponential time se-
ries is not enhanced by having more data points. Perception & Psychophysics,
24(2):182–184, ISSN: 0031-5117, 1532-5962, DOI: 10.3758/BF03199548, http:
//link.springer.com/10.3758/BF03199548.
Waldner, M., Diehl, A., Gračanin, D., Splechtna, R., Delrieux, C., and Matković, K.
(2020). A Comparison of Radial and Linear Charts for Visualizing Daily Patterns.
IEEE Transactions on Visualization and Computer Graphics, 26(1):1033–1042, ISSN:
2160-9306, DOI: 10.1109/TVCG.2019.2934784.
Washington Post (2022a). Coronavirus. https://www.washingtonpost.com/
coronavirus/.
Washington Post (2022b). U.S. coronavirus cases: Tracking deaths, con-
firmed cases by state. https://www.washingtonpost.com/nation/2020/03/27/
us-coronavirus-cases-tracking-deaths-confirmed-cases-by-state/.
Weiss, Y. (2020). These 12 Graphs Show Mask Mandates Do Noth-
ing To Stop COVID. https://thefederalist.com/2020/10/29/
these-12-graphs-show-mask-mandates-do-nothing-to-stop-covid/.
Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag
New York, ISBN: 978-3-319-24277-4, https://ggplot2.tidyverse.org.
World Health Organization (2022). WHO Coronavirus (COVID-19) Dashboard. https:
//covid19.who.int.
Yalcin, M. (2022). Mapping the global spatio-temporal dynamics of COVID-
19 outbreak using cartograms during the first 150 days of the pan-
demic. Geocarto International, 37(13):3791–3800, ISSN: 1010-6049, DOI:
10.1080/10106049.2020.1844310, https://doi.org/10.1080/10106049.2020.
1844310.
32 Visual narratives of COVID-19
Affiliation:
Adalbert F.X. Wilhelm
School for Business, Social and Decision Sciences
Jacobs University Bremen
Campus Ring 1
28759 Bremen
Germany
e-mail:a.wilhelm@jacobs-university.de
https://jdssv.org/
published by the International Association for Statistical Computing
http://iasc-isi.org/
MMMMMM YYYY, Volume VV, Issue II Submitted: yyyy-mm-dd
doi:XX.XXXXX/jdssv.v000.i00 Accepted: yyyy-mm-dd
View publication stats